In this paper, some new techniques for determining the observer-based or LQ
G form of any compensator with arbitrary order are discussed. The practical
appeal of such techniques is that they allow for a simplified implementati
on and reduced memory storage of general controllers and offer additional f
lexibility for handling gain-scheduling and input saturation constraints as
compensator states become meaningful variables. The derived observed-based
controllers are input-output equivalent to the original controller but wit
h an explicit separated estimation/control structure. Such structures invol
ve both static control and estimation gains with an extra Youla parameter t
hat can be either static or dynamic. The proposed techniques are applicable
both in continuous and discrete time, to full-order controllers, that is,
controllers whose order is the same as the plant's order but also to augmen
ted- and reduced-order controllers whose orders are greater or smaller, res
pectively. Necessary conditions to apply this general controller equivalenc
e principle are derived.
The interest and practicality of such techniques are then investigated with
regards to the LQG implementation of H-infinity and mu controllers, classe
s or controllers that do not generally enjoy ease of implementations. (C) 1
999 John Wiley & Sons, Ltd.